Building Low-Code AI Agent Flows with Flowise on the TrueFoundry AI Gateway

Over the past few months we’ve watched low-code builders like Flowise have grown in popularity Their drag-and-drop canvas lets data scientists—and increasingly product managers—link prompts, tools, vector search and multi-step agents without writing a single line of Python. That speed is addictive; proof-of-concepts pop up in hours instead of weeks. Yet the moment those prototypes start adding real value, platform teams confront a hidden tax: every block in the Flowise canvas talks to a different model endpoint, carries its own API key, logs usage in a separate portal and lands on a different line of the company credit card. Repeat that across ten teams and dozens of experiments and suddenly no one can answer the three questions every CIO cares about: Who called what? How much did it cost? Was it safe?‍

TrueFoundry’s AI Gateway exists precisely to answer those questions. Today it processes more than a million LLM calls each day for organisations such as NVIDIA, CVS Health and Siemens, applying project-level authentication, per-request cost ceilings, latency SLOs and full audit trails—whether the request targets GPT-4o, Claude 3 or an in-house fine-tune. When customers realised their Flowise agents were slipping outside that safety net, they asked if the same guardrails could apply “out of the box.” The demand was clear, so we wired Flowise into the Gateway. No separate partnership, no new console—just a seamless extension of the controls enterprises already rely on. Many users never notice the governance gap until invoices spike or compliance rings the bell; this integration closes that gap before it can open.

Why low-code agents belong behind the Gateway‍

Flowise, an open-source builder inspired by LangChain, lets anyone stitch together prompts, tools and vector stores with a visual canvas. That power becomes even more useful when every request it makes is automatically:

• authenticated via project-scoped tokens
• routed to the right model provider
• logged for latency, cost and safety evaluation and
• capped by enterprise-wide budget limits.

The Gateway already delivers those controls for mission-critical traffic (> 1 million LLM calls per day). Adding Flowise means the governance you rely on for GPT-4o or Claude 3 extends, unchanged, to your low-code experiments.

For a quick look at Truefoundry's AI Gateway visit: Link

A quick look at the integration

Prerequisites

• A Flowise deployment (self-hosted or SaaS)
• A TrueFoundry personal-access token
• Your Gateway base URL (https://<your-org>.truefoundry.cloud/api/llm)

Set-up in two short steps

1. Paste your credentials into Flowise
2. Open Flowise → Credentials → choose “OpenAI Custom”.
3. Drop in the Gateway URL and your token—no wrapper code required.
4. Build an agent and point it to your model
5. Inside the canvas, add an AgentFlow node, connect it from Start, and paste the model-ID you copied from the TrueFoundry dashboard. Click Save; every LLM call now flows through the Gateway.
   

You can find a more detailed walk-through, complete with screenshots, in our docs

What you gain out of the box

Once those two fields are filled, Flowise inherits everything the Gateway already does for production workloads:

• Unified observability – token counts, p50/p95 latency and full trace replay in one place.
• Cost governance – per-team spending caps and auto alerts for runaway agents.
• Vendor choice – swap Anthropic for Mistral by changing a single drop-down, no canvas edits.
• Security & compliance – violations blocked at the edge, audit logs stored automatically.

Looking ahead

Today’s release focuses on making experimentation painless. Over the next couple of sprints we plan to:

• surface TrueFoundry model IDs inside the Flowise node picker,
• ship an evaluation template so you can A/B test flows without leaving the dashboard, and
• open up a shared Slack channel for early feedback and troubleshooting.